Distributed Concurrent Programs as Strategies in Games

Abstract

This paper models synchronized distributed computing by extending the games in [Nerode-Yakhnis-Yakhnis, [17]]. The intention is to provide a precise tool for reasoning about the correctness of distributed synchronized programs. [17] introduced the paradigm that concurrent programs denote state strategies in corresponding two-person computational games. Program specifications were winning sets of plays in such games. This was carried out relative to a shared memory model. It is our belief that state strategies on appropriate game trees capture all there is to any program. So here we extend this paradigm to synchronous message-passing models, which may have shared memory as well, by introducing additional rules of the game, and additional operations for constructing new strategies from old, which capture completely the syntax and semantics of OCCAM’S PAR program construct and message-passing. Almost all the results about operations on strategies and programs from [17] extend to the distributed context.